A wind farm wake modeling and power generation prediction method, device, equipment, storage medium and program product

Abstract

The wind farm wake modeling and power generation prediction method, device, equipment, storage medium and program product provided by the embodiments of the present disclosure comprise: extracting terrain features and airflow distribution rules from multi-source data through an analysis sub-model in a wake proxy model, and adjusting the airflow distribution rules to obtain wake distribution rules according to unit layout data; establishing a wind farm graph structure according to the wake distribution rules, terrain features and unit layout data through a construction sub-model in the wake proxy model; mapping the wind farm graph structure into a prediction output vector set containing reference wind speed and power generation power through a double-encoding network and multi-layer message passing in a prediction sub-model in the wake proxy model; and efficiently capturing the spatial propagation characteristics of the wake influence between units through the double-encoding network and multi-layer message passing, so as to ensure the prediction accuracy while taking into account the calculation efficiency, and meet the rapid evaluation demand of multiple working conditions.

Description

Technical Field

[0001] This disclosure relates to the field of wind farm wake modeling technology, and in particular to a method, device, equipment, storage medium and program product for wind farm wake modeling and power generation prediction. Background Technology

[0002] The wake effect in wind farms is a significant factor affecting the power generation performance of wind turbines. Currently, wake modeling methods mainly include analytical wake surrogate models and computational fluid dynamics (CFD) numerical simulation methods. Analytical models have high computational efficiency but struggle to accurately describe the wake evolution process under terrain influence. CFD methods that consider terrain and multi-unit coupling can more accurately characterize the wind farm flow field and wake characteristics, but their computational cost is high, making it difficult to meet the needs of rapid assessment under multiple operating conditions in engineering applications. In recent years, data-driven wake surrogate models have gradually developed, but existing methods often fail to consider terrain factors or rely entirely on high-precision numerical simulation data, resulting in high computational costs and making it difficult to balance computational efficiency and prediction accuracy. Summary of the Invention

[0003] To solve the above-mentioned technical problems, or at least partially solve them, this disclosure provides a method, apparatus, equipment, storage medium, and program product for wind farm wake modeling and power generation prediction.

[0004] This disclosure provides a method for wind farm wake modeling and power generation prediction, the method comprising: Acquire the incoming flow data, turbine layout data, and terrain data of the target wind farm. Input the incoming flow data, turbine layout data, and terrain data into the analysis sub-model in the preset wake proxy model to determine the terrain features and airflow distribution patterns respectively. Based on the analysis sub-model, adjust the airflow distribution patterns according to the turbine layout data to obtain the wake distribution patterns. The wake proxy model includes an analysis sub-model, a construction sub-model, and a prediction sub-model. The terrain features, the turbine layout data, and the wake distribution pattern are input into the construction sub-model. Based on the construction sub-model, the global feature vector, node set, edge set, edge feature matrix, and node feature matrix of the target wind farm are determined. The wind farm map structure of the target wind farm is established according to the global feature vector, node set, edge set, edge feature matrix, and node feature matrix. The wind farm map structure is input into the prediction sub-model, and the prediction output vector set that maps the wind farm map structure to the target wind farm is generated based on the prediction sub-model. The prediction output vector set includes reference wind speed and power generation.

[0005] The method provided in this disclosure acquires incoming flow data, turbine layout data, and terrain data of a target wind farm. The incoming flow data, turbine layout data, and terrain data are input into an analysis sub-model within a preset wake proxy model. Terrain features and airflow distribution patterns are determined respectively. Based on the analysis sub-model and the turbine layout data, the airflow distribution patterns are adjusted to obtain the wake distribution patterns, including: The terrain data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the terrain elevation of the target wind farm; The unit layout data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the location elevation and airflow distribution influence factors of the target wind farm. The slope amplitude of the target wind farm is calculated based on the terrain elevation and the location elevation. The average slope is calculated based on the slope amplitude. The terrain complexity of the target wind farm is determined by comparing the average slope with a preset grading standard. The terrain features are determined by combining the slope amplitude, average slope, and terrain complexity. The airflow distribution pattern of the target wind farm is determined by analyzing the incoming flow data based on the analysis sub-model. The airflow distribution law is adjusted according to the airflow distribution influencing factor to obtain the wake distribution law.

[0006] The method provided in this disclosure embodiment inputs the terrain features, the turbine layout data, and the wake distribution pattern into the construction sub-model, determines the global feature vector, node set, edge set, edge feature matrix, and node feature matrix of the target wind farm based on the construction sub-model, and establishes the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix, and node feature matrix, including: Based on the constructed sub-model, the nodes and node feature vectors corresponding to each unit in the unit layout data are determined according to the terrain features. Based on the constructed sub-model, the mutual influence relationship between units is determined according to the wake distribution law and the node feature vector, and the edge and edge feature vector of each unit is determined according to the mutual influence relationship; The set of all nodes is the node set, the set of all edges is the edge set, the set of all edge eigenvectors is the edge feature matrix, and the set of all node eigenvectors is the node feature matrix. A global feature vector is determined based on the terrain features and the wake distribution pattern. The wind farm graph structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix.

[0007] The method provided in this disclosure, which inputs the wind farm map structure into the prediction sub-model and maps the prediction output vector set of the wind farm map structure to the target wind farm based on the prediction sub-model, includes: The prediction sub-model includes a first coding network and a second coding network; Based on the first encoding network, the node feature matrix and the global feature vector are mapped to a first mapping set, which includes the first mapping quantity corresponding to all nodes in the wind farm diagram structure. Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, which includes the second mapping quantities corresponding to all edges in the wind farm graph structure. The predicted output vector set of the target wind farm is determined based on the first mapping value and the second mapping value.

[0008] The method provided in this disclosure, based on the first encoding network, maps the node feature matrix and the global feature vector to a first mapping set, including: Based on the first encoding network, the node feature vector corresponding to the target node in the node feature matrix and the global feature vector are mapped from the target physical quantity space to a potential space with at least two layers to obtain the first mapping quantity of the target node in the first layer of the potential space. The first mapping set is obtained by combining the first mappings of all nodes.

[0009] The method provided in this disclosure, based on the second coding network, maps the edge feature vectors to a second mapping set, and determines the predicted output vector set of the target wind farm according to the first mapping amount and the second mapping amount, including: Based on the second encoding network, the edge feature vector corresponding to the target node is mapped from the target physical quantity space to the potential space to obtain the second mapping quantity of the two nodes associated with the edge corresponding to the edge feature vector; Based on the first mapping quantity and the second mapping quantity corresponding to the two nodes associated with the edge in the first layer, the message intermediate quantity of the first layer is determined. Aggregate the intermediate message quantities of the upstream neighboring nodes corresponding to the target node to obtain an aggregated message, and determine the first mapping quantity of the second layer based on the first mapping quantity of the aggregated message and the target layer; Complete message passing across all layers, and extract the first mapping of the target node in the last layer as the node's potential representation; The latent representation of the node is mapped back to the target physical quantity space to obtain the predicted output vector of the target node; The second mapping set is obtained by taking the second mapping values ​​corresponding to all edge feature vectors; The predicted output vector set is obtained by combining the predicted output vectors corresponding to all nodes in the target wind farm.

[0010] This disclosure also provides a wind farm wake modeling and prediction device, the device comprising: The analysis module is used to acquire the incoming flow data, turbine layout data, and terrain data of the target wind farm. The incoming flow data, turbine layout data, and terrain data are input into the analysis sub-model in the preset wake proxy model to determine the terrain features and airflow distribution patterns. Based on the analysis sub-model, the airflow distribution patterns are adjusted according to the turbine layout data to obtain the wake distribution patterns. The wake proxy model includes an analysis sub-model, a construction sub-model, and a prediction sub-model. The construction module is used to input the terrain features, the turbine layout data and the wake distribution pattern into the construction sub-model, determine the global feature vector, node set, edge set, edge feature matrix and node feature matrix of the target wind farm based on the construction sub-model, and establish the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix and node feature matrix; The prediction module is used to input the wind farm map structure into the prediction sub-model, and based on the prediction sub-model, to map the wind farm map structure into a prediction output vector set that represents the target wind farm. The prediction output vector set includes reference wind speed and power generation.

[0011] The analysis module in the apparatus provided in this disclosure is specifically used for: The terrain data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the terrain elevation of the target wind farm; The unit layout data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the location elevation and airflow distribution influence factors of the target wind farm. The slope amplitude of the target wind farm is calculated based on the terrain elevation and the location elevation. The average slope is calculated based on the slope amplitude. The terrain complexity of the target wind farm is determined by comparing the average slope with a preset grading standard. The terrain features are determined by combining the slope amplitude, average slope, and terrain complexity. The airflow distribution pattern of the target wind farm is determined by analyzing the incoming flow data based on the analysis sub-model. The airflow distribution law is adjusted according to the airflow distribution influencing factor to obtain the wake distribution law.

[0012] The apparatus provided in this disclosure, wherein the construction module is specifically used for: Based on the constructed sub-model, the nodes and node feature vectors corresponding to each unit in the unit layout data are determined according to the terrain features. Based on the constructed sub-model, the mutual influence relationship between units is determined according to the wake distribution law and the node feature vector, and the edge and edge feature vector of each unit is determined according to the mutual influence relationship; The set of all nodes is the node set, the set of all edges is the edge set, the set of all edge eigenvectors is the edge feature matrix, and the set of all node eigenvectors is the node feature matrix. A global feature vector is determined based on the terrain features and the wake distribution pattern. The wind farm graph structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix.

[0013] The apparatus provided in this disclosure, wherein the prediction module is specifically used for: The prediction sub-model includes a first coding network and a second coding network; Based on the first encoding network, the node feature matrix and the global feature vector are mapped to a first mapping set, which includes the first mapping quantity corresponding to all nodes in the wind farm diagram structure. Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, which includes the second mapping quantities corresponding to all edges in the wind farm graph structure. The predicted output vector set of the target wind farm is determined based on the first mapping value and the second mapping value.

[0014] The apparatus provided in this disclosure, wherein the prediction module is specifically used for: Based on the first encoding network, the node feature vector corresponding to the target node in the node feature matrix and the global feature vector are mapped from the target physical quantity space to a potential space with at least two layers to obtain the first mapping quantity of the target node in the first layer of the potential space. The first mapping set is obtained by combining the first mappings of all nodes.

[0015] The apparatus provided in this disclosure, wherein the prediction module is specifically used for: Based on the second encoding network, the edge feature vector corresponding to the target node is mapped from the target physical quantity space to the potential space to obtain the second mapping quantity of the two nodes associated with the edge corresponding to the edge feature vector; Based on the first mapping quantity and the second mapping quantity corresponding to the two nodes associated with the edge in the first layer, the message intermediate quantity of the first layer is determined. Aggregate the intermediate message quantities of the upstream neighboring nodes corresponding to the target node to obtain an aggregated message, and determine the first mapping quantity of the second layer based on the first mapping quantity of the aggregated message and the target layer; Complete message passing across all layers, and extract the first mapping of the target node in the last layer as the node's potential representation; The latent representation of the node is mapped back to the target physical quantity space to obtain the predicted output vector of the target node; The second mapping set is obtained by taking the second mapping values ​​corresponding to all edge feature vectors; The predicted output vector set is obtained by combining the predicted output vectors corresponding to all nodes in the target wind farm.

[0016] This disclosure also provides an electronic device, the electronic device comprising: a processor; a memory for storing executable instructions of the processor; the processor being configured to read the executable instructions from the memory and execute the instructions to implement the wind farm wake modeling and power generation prediction method provided in this disclosure.

[0017] This disclosure also provides a computer-readable storage medium storing a computer program for executing the wind farm wake modeling and power generation prediction method provided in this disclosure.

[0018] The technical solution provided in this disclosure has the following advantages compared with the prior art: The wind farm wake modeling and power generation prediction method provided in this embodiment extracts terrain features and airflow distribution patterns from multi-source data through the analysis sub-model in the wake proxy model, and adjusts the airflow distribution patterns according to the turbine layout data to obtain the wake distribution patterns. The construction sub-model in the wake proxy model establishes the wind farm map structure based on the wake distribution patterns, terrain features, and turbine layout data. Through the dual-encoding network and multi-layer message passing in the prediction sub-model in the wake proxy model, the wind farm map structure is mapped into a prediction output vector set containing reference wind speed and power generation. The dual-encoding network and multi-layer message passing efficiently capture the spatial propagation characteristics of wake influence between turbines, balancing computational efficiency with prediction accuracy, and meeting the needs of rapid evaluation of multi-condition engineering projects. Attached Figure Description

[0019] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0020] Figure 1A flowchart illustrating the wind farm wake modeling and power generation prediction method provided in this embodiment of the disclosure; Figure 2 A schematic diagram of the wind farm wake modeling process provided in this embodiment of the disclosure; Figure 3 This is a schematic diagram for calculating the average slope. Figure 4 (a) is a schematic diagram of the wind farm structure provided in an embodiment of this disclosure; Figure 4 (b) in the figure is a schematic diagram of the prediction output vector provided in the embodiments of this disclosure; Figure 5 A schematic diagram of the prediction sub-model framework provided in the embodiments of this disclosure; Figure 6 This is a schematic diagram showing the location of the wind measurement tower and turbine units in a wind farm example provided in this embodiment of the disclosure; Figure 7 (a) in the figure is a wind speed distribution diagram of the wind measurement tower provided in the embodiment of this disclosure; Figure 7 (b) is a turbulence intensity distribution diagram of the wind measurement tower provided in the embodiments of this disclosure; Figure 8 Power and thrust coefficient curves of a wind turbine provided in the embodiments of this disclosure; Figure 9 Flow condition distribution diagrams for multi-precision training and verification scenarios provided in this embodiment of the present disclosure; Figure 10 (a) in the figure is a schematic diagram comparing the reference wind speed and the measured wind speed in the predicted output vectors of the three methods under the first incoming flow condition; Figure 10 (b) in the figure is a schematic diagram comparing the power generation and the measured power generation in the predicted output vectors of the three methods under the first incoming flow condition; Figure 10 (c) in the figure is a schematic diagram comparing the reference wind speed and the measured wind speed in the predicted output vectors of the three methods under the second incoming flow condition in the example. Figure 10 (d) in the figure is a schematic diagram comparing the power generation and the measured power generation in the predicted output vector of the three methods under the second incoming flow condition in the example. Figure 11 A schematic diagram of the wind farm wake modeling and prediction device provided in an embodiment of this disclosure; Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, the method embodiments may include additional steps and / or omit the steps shown. The scope of this disclosure is not limited in this respect.

[0023] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0024] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0025] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0026] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0027] To address the aforementioned issues, this disclosure provides a method for wind farm wake modeling and power generation prediction. The method will be described below with reference to specific embodiments.

[0028] Figure 1 This is a flowchart illustrating a wind farm wake modeling and power generation prediction method provided in an embodiment of the present disclosure. The method can be executed by a wind farm wake modeling and prediction device, which can be implemented in software and / or hardware and is generally integrated into an electronic device.

[0029] Example 1: This embodiment of the present disclosure provides a method for wind farm wake modeling and power generation prediction, such as... Figure 1 and Figure 2 As shown, it includes: S101: Obtain the incoming flow data, turbine layout data, and terrain data of the target wind farm. Input the incoming flow data, turbine layout data, and terrain data into the analysis sub-model in the preset wake proxy model, determine the terrain features and airflow distribution patterns respectively, and adjust the airflow distribution patterns according to the turbine layout data based on the analysis sub-model to obtain the wake distribution pattern. The wake proxy model includes an analysis sub-model, a construction sub-model, and a prediction sub-model. S102: Input the terrain features, the turbine layout data and the wake distribution pattern into the construction sub-model, determine the global feature vector, node set, edge set, edge feature matrix and node feature matrix of the target wind farm based on the construction sub-model, and establish the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix and node feature matrix; S103: Input the wind farm map structure into the prediction sub-model, and based on the prediction sub-model, map the wind farm map structure into the prediction output vector set of the target wind farm. The prediction output vector set includes reference wind speed and power generation.

[0030] In this embodiment, the incoming flow data refers to the wind speed, direction, turbulence intensity, and wind profile parameters of the free-flowing air. Its function is to provide inflow boundary conditions and drive the generation of airflow distribution patterns. The unit layout data includes unit coordinates, altitude, hub height, and thrust coefficient curves. Its function is to provide geometric and operational parameters for upstream and downstream mapping, node definition, and wake superposition. The terrain data is a high-resolution digital elevation model (DEM) raster. Its function is to extract elevation, average slope, and aspect, calculate terrain acceleration and deceleration effects, and form a terrain feature field, where, for example... Figure 3 The diagram illustrates the calculation process for the average slope. The average slope is a statistical measure obtained by arithmetically averaging all slope amplitudes across the entire wind farm or within a specific section. Its purpose is to reflect the overall tilt of the wind farm with a single numerical value, providing a quantitative basis for classifying terrain complexity. The corresponding calculation formula is... Where K is the number of valid sampling points; This represents the average slope of the area to be laid out. This represents the slope magnitude of the i-th sampling point relative to the unit's location. ,in, Let be the terrain elevation of the i-th sampling point; The elevation of the unit's location is indicated by the slope amplitude; the slope amplitude refers to the arctangent of the ratio of the elevation difference between adjacent terrain features to the horizontal distance, representing the degree of local surface inclination. Its function is to provide the basic statistical unit for calculating the average slope. The spatial distribution of the slope amplitude reflects the severity of terrain undulation in the site area. The wake surrogate model is an overall framework composed of an analysis sub-model, a construction sub-model, and a prediction sub-model, and its function is to calculate the annual power generation of each candidate unit layout. The analysis sub-model is responsible for multi-source data parsing and physical characteristic extraction. Its function is to determine terrain features from terrain data, determine airflow distribution patterns from incoming flow data, and then adjust the wake distribution pattern according to the unit layout. Terrain features are spatial parameters such as slope, aspect, undulation, and acceleration / deceleration factors extracted from the DEM. Their function is to quantify geographical undulations into numerical correction parameters for use in airflow distribution adjustment and node feature construction. The airflow distribution pattern is the initial spatial distribution of wind speed across the entire site obtained based on incoming flow data and terrain features. Its function is to provide an environmental wind field benchmark before the superposition of multi-unit wake interference. The adjustment process involves establishing upstream and downstream mappings based on the turbine layout and applying wake superposition and deflection models to correct airflow distribution. Its purpose is to transform the ideal wake of a single turbine into a coupled wake of multiple turbines. The wake distribution pattern, determined after turbine layout adjustments, represents the final spatial distribution of wind speed and turbulence across the entire field. Its function is to fully describe the wake velocity deficit and turbulence enhancement characteristics, providing physical field information for constructing sub-models. The sub-model receives terrain features, turbine layout, and wake distribution patterns, and outputs a wind farm map structure. Its function is to transform physical field information into graph data and establish the wind farm map structure. The global feature vector is a matrix fusing all node attributes and environmental parameters. Its function is to map the wind farm map structure to the initial potential representation of each node via the first encoding network. The node set is the complete set of nodes abstracted from all turbines. Its function is to provide computational units for the graph structure; message passing and representation updates are performed on a node-by-node basis. The edge set is the complete set of directed edges determined based on wind direction and wake geometry. Its function is to define the topological paths of wake influence between turbines, providing propagation channels for message construction and aggregation. The edge feature matrix records the wake influence intensity and spatial relationship of each pair of directed edges, encoding the wake overlap area and relative distance. After mapping by the second encoding network, it participates in message construction. The node feature matrix stores the coordinates, elevation, inflow conditions, and terrain correction factor of each node, providing input to the first encoding network and generating the initial potential representation of the node at each layer. The wind farm graph structure is a directed graph consisting of a set of nodes, a set of edges, node feature matrices, and edge feature matrices, as shown below. Figure 4 As shown in (a), its function is to abstract the physical topology of a wind farm into a data structure that can be processed by a graph neural network, providing data support for the prediction sub-model. The wind farm graph structure vector is... ,in, For the reason The set of nodes formed by typhoon turbine units. Let n be the set of edges describing the interrelationships between units, where n represents the number of nodes. Figure 4 In (a), i and j represent node indices. Represents the node feature vector. , The dimension representing the node features. Representing the node feature matrix , Represents the edge feature vector. , Let the feature dimension be denoted as . Representing the edge feature matrix k represents the number of edges, and the edge feature vector also includes the relative distance. Relative position and the angle with the direction of the incoming flow , Represents the global feature vector. , This represents the global feature dimension, including inflow wind speed. ,wind direction and turbulence intensity The prediction sub-model receives the wind farm map structure and outputs a prediction vector set. , Represents the predicted output vector. , This indicates the dimension of the predicted output vector. The summation function represents the entire mapping process, and the predicted output vector includes the effective incoming wind speed at unit i. and the corresponding unit output power ,like Figure 5 As shown, the mapping process is as follows: for any node and edge Its initial node potential representation And edge potential representation Defined as and ,in, This represents the first mapping value at level 0. The function representing the mapping process of the feature vector of the i-th node. The function representing the mapping process of global feature vectors. The second mapping quantity represents the edge feature vector between the i-th node and the j-th node. This represents the mapping process function of the edge feature vectors between the i-th node and the j-th node. Both the first and second mapping processes reside in the encoder. In the latent space, nodes interact with their neighboring nodes through a message-passing mechanism, which resides in the processor. In the layer, the message construction process between node i and its neighboring node j is represented as follows: , This represents the intermediate message quantity of node i at level l, node j. This represents the message construction process function for the l-th layer. This represents the first mapping value of the i-th node in the l-th layer. This represents the first mapping of the j-th node in the l-th layer, and the mapping of the i-th node to its neighbors. message aggregation ,in, This represents the aggregated message at layer l. Represents the aggregation function and updates the latent representation of the nodes accordingly. , The update function for layer l represents the node representation that, through multi-layer message passing, progressively integrates unit information from a larger spatial range, thereby learning the spatial propagation characteristics of wake effects between units. After completing layer L message passing, the final latent representation of the node is mapped back to the target physical quantity space. ,in, This is the decoding network, located in the decoder, used to output the reference wind speed and power prediction results for the corresponding units at each node. It includes a dual-encoding network and a multi-layer message passing mechanism, its function being to map the graph structure to a latent space representation and determine the prediction output vector, such as... Figure 4 (b) Reference wind speed and power generation. The mapping process involves mapping the graph structure to a multi-layer latent space through a dual-encoding network, performing forward propagation of message construction, aggregation, and node updates layer by layer. Its function is to transform the original graph data in the wind farm graph structure into a node latent representation that incorporates multi-hop wake information.

[0031] To verify the effectiveness of the embodiments of the present invention, an actual wind farm was selected as an implementation case to illustrate the multi-precision wind farm wake proxy modeling method considering the influence of terrain. The selected wind farm is located in an area combining plains and gentle hills, with some terrain undulations. The elevation range of the site is approximately 70–170 m, and the area is approximately 80 km². The turbulence intensity is moderate, and the wake effect is significant. The wind farm terrain and the location of the wind turbines are as follows: Figure 6 As shown in Table 1, the wind turbine hub height is 90 m, the rotor diameter is 103 m, and the rated power is 1.6 MW. The low-precision training inflow condition parameter settings are shown in Table 1. Table 1. Low-precision training incoming flow condition parameter settings

[0032] Subsequently, seven turbine units were selected for verification. Measured data from the anemometer towers were used as the inflow conditions, and the model-predicted wake and power results were compared and analyzed using SCADA data from the units. Based on the anemometer tower measured data, the wind speed and turbulence intensity distribution characteristics of the study area were statistically analyzed, and the results are as follows: Figure 7 (a) and Figure 7 As shown in (b). Based on the statistical results and the thrust coefficient curve of the wind turbine, the incoming wind speed range was set to 3–15 m / s, and the environmental turbulence intensity range was set to 0.02–0.30. Based on the analytical wake model, the incoming flow conditions were discretized at certain intervals within the above parameter range to generate multiple sets of sample operating conditions. For each set of operating conditions, the reference wind speed under the influence of the wake of each turbine was calculated; combined with the wind turbine power curve, the output power of the interpolation computer was used to construct a low-precision wake database. The turbine power curve is shown in... Figure 8 As shown in Table 2, the wind speed at a position 2.5 times the rotor diameter in front of each unit's rotor is extracted as the reference wind speed, and the unit power is calculated in conjunction with the power curve. Representative CFD simulation conditions are selected based on the incoming flow range; the incoming flow conditions are shown in Table 2.

[0033] Table 2. Flow condition parameter settings for CFD simulation conditions

[0034] Operating conditions with wind direction of 80-100° and wind speed and environmental turbulence intensity falling within the above parameter range were selected from the measured samples. A verification sample set was constructed using SCADA data from seven units in the field, and its parameter settings are shown in Table 3. The inflow conditions distribution for the multi-precision training and verification operating conditions is as follows: Figure 9 As shown.

[0035] Table 3. Flow condition parameter settings for the verification working condition.

[0036] During the terrain feature calculation process, terrain information is extracted from the locations of each wind turbine based on the digital elevation model of the wind farm area. The elevation of the turbine locations is then used as the basis for the calculation. Based on this, terrain sampling is performed in the direction of the incoming flow, with a sampling distance of 5H. For distances of... The sampling points are used to calculate the slope magnitude relative to the unit's location. The average slope along the direction of the incoming flow is obtained by averaging all valid sampling points. . In the model construction process, the wind farm is abstracted as a graph structure, where each wind turbine corresponds to a node in the graph, and the spatial relationships between turbines constitute the connecting edges between nodes. Edge features are constructed by calculating the distance between turbines, their relative orientation, and the angle with the incoming flow direction. Simultaneously, the calculated terrain features are used as node feature inputs, while incoming wind speed, wind direction, and turbulence intensity are used as global feature inputs to the model. Based on this, a wake proxy model based on a graph neural network is constructed. The encoding module maps node features, edge features, and global features to the latent space, and performs multi-layer information transmission on the graph structure, allowing each node to gradually integrate wake influence information from upstream turbines, thereby learning the spatial correlation of wake propagation between turbines. Finally, the decoding module outputs the reference wind speed and power generation of each turbine. During model training, the model is first pre-trained using low-precision data to learn the basic laws of wind farm wake distribution. Then, representative high-precision samples are selected to fine-tune the model to correct systematic biases under terrain influence conditions, thereby improving overall prediction accuracy.

[0037] Based on a database of 16 validation operating conditions, CFD numerical simulation, GNN model, and GNN-T model were used to predict wind speed and power at the target turbine location under different operating conditions. The relative percentage error (MAPE) was used to evaluate the model's prediction performance for wind speed and power for each turbine in the validation sample. For the i-th turbine in sample k, the relative errors for wind speed and power can be expressed as follows: and ,in, This represents the relative error of wind speed for the i-th unit in the k-th sample. This represents the relative power error of the i-th unit in the k-th sample; This represents the predicted average reference wind speed of the i-th unit under the k-th sample; This represents the predicted average power generation of the i-th unit under the k-th sample; This represents the measured wind speed of the i-th unit in the k-th sample; Let represent the average power of the i-th unit under the k-th sample. Taking two typical inflow conditions (wind speed 6.69 m / s, turbulence intensity 0.14; wind speed 8.17 m / s, turbulence intensity 0.14) as examples, the MAPE at each unit location is calculated using the three methods, as shown in Table 4.

[0038] Table 4. Prediction errors of the three wake simulation methods at the target position. a) U0 = 6.69 m / s TI = 0.14

[0039] b) U0 = 8.17 m / s TI = 0.14

[0040] In addition, to more intuitively compare the prediction performance of different methods, a comparison chart of the predicted values ​​and measured values ​​for each method was drawn, and the results are as follows: Figure 10 As shown.

[0041] In summary, the GNN model has shown some predictive ability at some turbine locations, but its error fluctuations are large and its stability is insufficient. The CFD method yields relatively stable results, but it is not optimal at all turbine locations. The GNN-T model not only reflects the distribution patterns of wind speed and power among different turbine units well, but also exhibits higher prediction accuracy under two typical operating conditions. This indicates that after fine-tuning with high-precision data, the model's ability to represent the impact of turbine wakes and the differences between turbine units has been improved. To evaluate the computational efficiency of different wake models in engineering applications, the average computation time of the CFD numerical simulation method, the Niayifar_Gauss analytical model, and the proposed GNN wake model were compared and analyzed. The results are shown in Table 5.

[0042] Table 5. Wake calculation costs of three wake simulation methods

[0043] The computational cost comparison results show that the GNN model has a significant computational efficiency advantage over the CFD method, reducing the computation time per run by approximately 10%. 4 The computation time is on the order of magnitude, meeting the needs of rapid assessment under multiple operating conditions in wind farms. Compared with the analytical wake model, the GNN model has a slightly longer computation time, but it is still within the range of seconds. Overall, the proposed GNN wake surrogate model achieves a good balance between computational efficiency and prediction accuracy, and has good engineering application value.

[0044] The working principle and beneficial effects of this embodiment are as follows: The analysis sub-model in the wake proxy model extracts terrain features and airflow distribution patterns from multi-source data, and adjusts the airflow distribution patterns according to the turbine layout data to obtain the wake distribution patterns. The construction sub-model in the wake proxy model establishes a wind farm map structure based on the wake distribution patterns, terrain features, and turbine layout data. Through the dual-encoding network and multi-layer message passing in the prediction sub-model of the wake proxy model, the wind farm map structure is mapped into a set of predicted output vectors containing reference wind speed and power generation. The dual-encoding network and multi-layer message passing efficiently capture the spatial propagation characteristics of wake influence between turbines, ensuring both computational efficiency and prediction accuracy, and meeting the needs of rapid evaluation of multi-condition engineering.

[0045] Example 2: The method provided in this embodiment of the present disclosure acquires incoming flow data, turbine layout data, and terrain data of a target wind farm. The incoming flow data, turbine layout data, and terrain data are input into an analysis sub-model within a preset wake proxy model. Terrain features and airflow distribution patterns are determined respectively. Based on the analysis sub-model and the turbine layout data, the airflow distribution patterns are adjusted to obtain the wake distribution patterns, including: The terrain data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the terrain elevation of the target wind farm; The unit layout data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the location elevation and airflow distribution influence factors of the target wind farm. The slope amplitude of the target wind farm is calculated based on the terrain elevation and the location elevation. The average slope is calculated based on the slope amplitude. The terrain complexity of the target wind farm is determined by comparing the average slope with a preset grading standard. The terrain features are determined by combining the slope amplitude, average slope, and terrain complexity. The airflow distribution pattern of the target wind farm is determined by analyzing the incoming flow data based on the analysis sub-model. The airflow distribution law is adjusted according to the airflow distribution influencing factor to obtain the wake distribution law.

[0046] In this embodiment, the specific process of analyzing terrain data involves loading the DEM raster file, parsing the affine transformation parameters, converting spatial coordinates into row and column numbers, reading pixel elevation values, and simultaneously calculating derived parameters such as slope and aspect. Its function is to extract the overall elevation and terrain factors, providing a basic data source for subsequent terrain feature quantification. Terrain elevation is the ground elevation value of each spatial point extracted from the DEM. Its function is to provide an overall elevation benchmark, calculate the slope amplitude in conjunction with the location elevation, and provide elevation input for terrain complexity determination. The specific process of analyzing the generator set layout data involves parsing the generator set coordinate table, extracting the latitude, longitude, altitude, and aircraft type parameters of each generator set, and determining the spatial arrangement and hub height of the generator sets. Its function is to obtain the precise attributes of each generator position, providing data support for location elevation extraction and calculation of airflow distribution influence factors. Location elevation is the ground elevation of each generator position directly read from the generator set layout data or sampled from the DEM. Its function is to calculate the slope amplitude in conjunction with the terrain elevation and provide generator position elevation attributes for subsequent node feature vector construction. The airflow distribution influencing factor is a set of wind speed correction coefficients calculated based on generator layout and terrain features, including terrain acceleration factors, deceleration factors, and wake superposition weights. Its function is to perform terrain-differentiated correction of airflow distribution patterns, ensuring that the airflow distribution reflects the actual coupling effect between terrain and generator. The process of comparing preset grading standards involves matching the average slope with a pre-set threshold range, typically categorized into flat, complex, and extremely complex levels. Its function is to convert continuous slope values ​​into discrete terrain levels, supporting the qualitative classification of terrain features. Terrain complexity is the terrain level classification result obtained based on the average slope and grading standards, typically categorized into flat, complex, and extremely complex terrain. Its function is to provide a basis for wake model selection and parameter setting, determining whether to enable the complex terrain correction module. Terrain features are a set of spatial parameters composed of slope amplitude, average slope, terrain complexity, and terrain acceleration / deceleration factors. Its function is to convert geographical undulations into a quantitative parameter field, providing complete input for adjusting airflow distribution patterns and constructing graph structure node features. The process of analyzing incoming flow data refers to statistically analyzing wind condition parameters such as wind rose diagrams, average wind speed, turbulence intensity, and wind profile indices based on anemometer or reanalysis data. Its function is to extract the inflow boundary conditions of the wind farm, drive the generation of initial airflow distribution patterns, and provide wind condition input for wake calculations.

[0047] The working principle and beneficial effects of this embodiment are as follows: Topographic elevation and average slope are extracted from topographic data through analysis of the sub-model, and topographic complexity is determined by combining the classification criteria, thereby obtaining topographic features. Airflow distribution patterns are obtained from incoming flow data, and then adjusted and corrected based on airflow distribution influencing factors derived from the unit layout, outputting wake distribution patterns. Slope and topographic complexity are calculated jointly using topographic elevation and location elevation, achieving quantitative extraction of complex topographic features. The initial airflow distribution is corrected using airflow distribution influencing factors, effectively integrating multi-aircraft wake interference effects, balancing computational efficiency and the prediction accuracy of wake distribution patterns.

[0048] Example 3: The method provided in this embodiment of the present disclosure inputs the terrain features, the turbine layout data, and the wake distribution pattern into the construction sub-model, determines the global feature vector, node set, edge set, edge feature matrix, and node feature matrix of the target wind farm based on the construction sub-model, and establishes the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix, and node feature matrix, including: Based on the constructed sub-model, the nodes and node feature vectors corresponding to each unit in the unit layout data are determined according to the terrain features. Based on the constructed sub-model, the mutual influence relationship between units is determined according to the wake distribution law and the node feature vector, and the edge and edge feature vector of each unit is determined according to the mutual influence relationship; The set of all nodes is the node set, the set of all edges is the edge set, the set of all edge eigenvectors is the edge feature matrix, and the set of all node eigenvectors is the node feature matrix. A global feature vector is determined based on the terrain features and the wake distribution pattern. The wind farm graph structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix.

[0049] In this embodiment, a node is a basic computational unit abstracted from each wind turbine in a graph structure. Its function is to provide a carrier for message passing and feature updates in the graph network. Each node independently encodes and aggregates neighborhood information, supporting the propagation and fusion of wake effects along the topology. The node feature vector is a numerical vector encoding the attributes of each turbine and the terrain features of its location, including coordinates, altitude, inflow conditions, and terrain correction factors. Its function is to serve as input to the first encoding network, generating the initial mapping of nodes in the latent space and driving subsequent message passing and representation updates. The mutual influence relationship is defined based on the upstream wind direction and wake geometry, indicating an upstream-downstream dependency between turbines. It defines the influence link of the upstream turbine's wake on the downstream turbine, which varies depending on the wind direction. The edge provides a topological basis for establishing the edge set, determining the direction and connected objects of edges in the graph structure. An edge is a directed connection from an upstream turbine to a downstream turbine, representing a propagation path with wake influence. Since wind directions are not uniform, the mutual influence relationship is used consistently. The first encoding network provides a propagation channel for message construction and aggregation. Messages flow along edges from upstream nodes to downstream nodes, enabling graph structure modeling of wake interference. Edge feature vectors are numerical vectors encoding the wake influence intensity and spatial relationship of each edge, including attributes such as wake overlap area, relative distance, and wind speed deficit. They participate in message construction after mapping through the second encoding network, quantifying the wake influence weight carried by each edge. The node set is the complete set of nodes abstracted from all units in the wind farm. It provides a complete list of computational nodes for the graph structure, defines the scope of message passing and representation updates, and supports iterative computation across the entire graph. The node feature matrix is ​​a two-dimensional matrix formed by stacking all node feature vectors in rows, with rows corresponding to unit nodes and columns corresponding to feature dimensions. It unifies discrete node features into structured tensors, serving as batch inputs to the first encoding network and improving computational efficiency. The edge set is the complete set of all directed edges. Together with the node set, it defines the topological skeleton of the graph structure, provides a complete connection table for message passing, and supports neighborhood aggregation operations. The edge feature matrix is ​​a two-dimensional matrix formed by stacking all edge feature vectors in rows. The function is to uniformly store edge attributes as batch inputs to the second encoding network, enabling parallel mapping of edge features to the latent space. The global feature vector is a summary representation of the overall characteristics, integrating the statistical values ​​of all node attributes and the wake distribution patterns. Its function is to provide each node with global contextual information about the wind farm; concatenated with the node feature vector, it is input into the first encoding network, enhancing the global perception capability of the node's initial mapping. The wind farm graph structure is a directed graph data entity composed of the node set, edge set, node feature matrix, edge feature matrix, and global feature vector. Its function is to unify the wind farm's physical topology and attribute information into a data structure that can be processed by graph neural networks, supporting end-to-end prediction.

[0050] The working principle and beneficial effects of this embodiment are as follows: By constructing a sub-model based on the unit layout data and combining terrain features, each unit is abstracted into a node and node feature vectors are generated; the mutual influence relationship between units is determined according to the wake distribution law and node features, and the edges and edge feature vectors are determined. By combining the nodes, edges, feature matrices and global feature vectors generated based on terrain and wake laws, a complete wind farm graph structure is constructed. The terrain features and wake distribution laws are uniformly transformed into graph structure data, realizing the system mapping of physical field information to graph topology. Through the independent construction of node features and edge features, the unit attributes and wake influence relationship are fully preserved, providing a structured input basis for subsequent graph neural network safe mapping.

[0051] Example 4: The method provided in this embodiment of the present disclosure inputs the wind farm map structure into the prediction sub-model, and maps the prediction output vector set of the wind farm map structure to the target wind farm based on the prediction sub-model, including: The prediction sub-model includes a first coding network and a second coding network; Based on the first encoding network, the node feature matrix and the global feature vector are mapped to a first mapping set, which includes the first mapping quantity corresponding to all nodes in the wind farm diagram structure. Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, which includes the second mapping quantities corresponding to all edges in the wind farm graph structure. The predicted output vector set of the target wind farm is determined based on the first mapping value and the second mapping value.

[0052] In this embodiment, the first encoding network is a multi-layer fully connected network structure responsible for node feature mapping, transforming the node feature matrix and global feature vector from the physical quantity space to the latent space. Its function is to generate an initial latent representation for each node, preserving the fusion information of node attributes and global context, and supporting the initialization of node states for subsequent message passing. The second encoding network is a multi-layer fully connected network structure responsible for edge feature mapping, transforming the edge feature matrix from the physical quantity space to the latent space. Its function is to generate a latent representation for each edge, encoding the intensity and spatial relationship of wake influence, and providing abstract feature inputs of edge attributes for message construction. The first mapping set is the set of first mapping quantities obtained after mapping all nodes in the wind farm graph structure through the first encoding network. The first mapping quantity is the vector representation of a single node in the latent space after mapping through the first encoding network, located in the initial layer of the latent space. Its function is to serve as the initial state of the node in message passing, participate in the first-layer message construction, and gradually fuse neighborhood wake information with each layer iteration. The second mapping set is the set of second mapping quantities obtained after mapping all edges in the wind farm graph structure through the second encoding network. The second mapping is the vector representation of a single edge in the latent space after being mapped by the second encoding network, encoding the strength of the wake influence carried by the edge. Its function is to fuse the first mappings of the two nodes associated with the edge, constructing the intermediate message quantity corresponding to that edge, and driving neighborhood information aggregation and node state updates.

[0053] The working principle and beneficial effects of this embodiment are as follows: A first encoding network maps node feature matrices and global feature vectors to a first mapping set, and a second encoding network maps edge feature matrices to a second mapping set. Based on the first and second mapping values, wake influence information is aggregated in a multi-hop neighborhood through a message passing mechanism, ultimately generating a predicted output vector set containing reference wind speed and power generation. The dual encoding network independently encodes node and edge features into the latent space, preserving the differentiated expression of unit attributes and wake influence relationships; multi-hop message passing captures the wake space propagation characteristics, achieving end-to-end mapping from graph structure to predicted output, balancing computational efficiency and prediction accuracy.

[0054] Example 5: The method provided in this embodiment of the present disclosure, based on the first encoding network, maps the node feature matrix and the global feature vector to a first mapping set, including: Based on the first encoding network, the node feature vector corresponding to the target node in the node feature matrix and the global feature vector are mapped from the target physical quantity space to a potential space with at least two layers to obtain the first mapping quantity of the target node in the first layer of the potential space. The first mapping set is obtained by combining the first mappings of all nodes.

[0055] In this embodiment, the target physical quantity space refers to the original feature dimension space where the node feature vector and the global feature vector originally reside, including a set of attributes with definite physical units such as unit coordinates, altitude, inflow wind speed, and terrain correction factors. Its function is to provide the representation domain of the original input data for the encoding network, serving as the starting space for mapping operations, forming a correspondence with the latent space, and clarifying the boundaries and directions of feature transformation. The latent space with at least two layers refers to the multi-layer abstract feature space at the output of the first encoding network, with each layer corresponding to a feature representation of different granularities, and the layers are linked together through message passing. Its function is to provide a progressively evolving environment for node representations, with the lower layers capturing local neighborhood information and the higher layers fusing multi-hop wake propagation effects to support hierarchical feature learning.

[0056] The working principle and beneficial effects of this embodiment are as follows: The first encoding network fuses the node feature vector of the target node with the global feature vector, obtains the first mapping quantity by mapping from the physical quantity space to the latent space and summarizes it into the first mapping set, and transforms the physical quantity into an abstract representation in the latent space by jointly encoding the node attributes and the global context, retaining the constraints of the overall information of the wind farm on the single node, and providing the initial node representation for information fusion for subsequent message transmission.

[0057] Example 6: The method provided in this embodiment of the present disclosure, based on the second coding network, maps the edge feature vectors to a second mapping set, and determines the predicted output vector set of the target wind farm according to the first mapping amount and the second mapping amount, including: Based on the second encoding network, the edge feature vector corresponding to the target node is mapped from the target physical quantity space to the potential space to obtain the second mapping quantity of the two nodes associated with the edge corresponding to the edge feature vector; Based on the first mapping quantity and the second mapping quantity corresponding to the two nodes associated with the edge in the first layer, the message intermediate quantity of the first layer is determined. Aggregate the intermediate message quantities of the upstream neighboring nodes corresponding to the target node to obtain an aggregated message, and determine the first mapping quantity of the second layer based on the first mapping quantity of the aggregated message and the target layer; Complete message passing across all layers, and extract the first mapping of the target node in the last layer as the node's potential representation; The latent representation of the node is mapped back to the target physical quantity space to obtain the predicted output vector of the target node; The second mapping set is obtained by taking the second mapping values ​​corresponding to all edge feature vectors; The predicted output vector set is obtained by combining the predicted output vectors corresponding to all nodes in the target wind farm.

[0058] In this embodiment, the intermediate message quantity refers to the intermediate vector constructed by fusing the first and second mapping quantities of two nodes connected by an edge through a message function during message passing. Its function is to combine the state information of upstream nodes with the wake influence strength carried by the edge to generate the transmission information on a single edge, serving as the input unit for aggregation operations. The aggregated message refers to the single vector obtained by the target node after performing aggregation operations such as summation or averaging on the intermediate message quantities from all upstream neighboring nodes. Its function is to compress multiple upstream wake influence information into a fixed-dimensional neighborhood summary representation, providing a comprehensive context for node state updates. Message passing refers to the iterative process of information interaction between nodes and neighboring nodes at each layer, including message construction, message aggregation, and node update. Its function is to propagate the wake influence along directed edges in multiple hops through layer-by-layer iteration, enabling the node representation to gradually integrate upstream unit information from a larger spatial range. The node latent representation refers to the target node feature vector extracted from the last latent space after completing message passing at all layers. Its function is to finally abstract the node that has integrated multi-hop wake propagation information, directly mapping it to the predicted value as the input of the output network. Mapping back to the target physical quantity space refers to the operation of transforming the latent representation of the final layer nodes from the latent space back to the original physical quantity space through the output network. Its function is to decode abstract features into predicted values ​​with clear physical meaning, completing the closed-loop transformation from graph structure to engineering prediction output. The output network refers to the fully connected layer structure connecting the final layer latent representation and the prediction output vector. Its function is to execute the inverse mapping function from the latent space to the physical quantity space, nonlinearly transforming the node latent representation into a prediction output of a specified dimension. The prediction output vector refers to the prediction result vector obtained after mapping a single target node through the output network, containing reference wind speed and power generation. Its function is to provide quantifiable wake effect prediction values ​​for a single unit, supporting single-unit performance evaluation. The prediction output vector set refers to the aggregated set of prediction output vectors for all target nodes in the entire wind farm. Its function is to provide complete prediction results for all units in the wind farm, supporting farm-level power generation statistics and unit operation status analysis.

[0059] The working principle and beneficial effects of this embodiment are as follows: The second encoding network maps the edge feature vectors to the latent space to obtain the second mapping quantity, which is then fused with the first mapping quantity to construct the intermediate message quantity. After aggregating neighborhood messages, the node representation is updated. After multi-layer transmission, the latent representation of the final layer node is extracted and mapped back to the physical quantity space to obtain the predicted output vector. These are then summarized to form a predicted output vector set. By independently encoding edge features through the second encoding network and cooperating with the node mapping quantity to construct messages, the differentiated expression of the wake effect intensity is achieved. Multi-layer message transmission aggregates the wake effect of upstream units hop by hop. The final layer node representation fuses multi-hop wake information and maps it back to the physical space to output wind speed and power, thereby improving prediction accuracy.

[0060] To achieve the above embodiments, this disclosure also proposes a wind farm wake modeling and prediction device.

[0061] Figure 11 This is a schematic diagram of the wind farm wake modeling and prediction device provided in an embodiment of this disclosure. The device 200 can be implemented by software and / or hardware, and is generally integrated into an electronic device. For example... Figure 11 As shown, the device 200 includes: an analysis module 201, a construction module 202, and a prediction module 203, wherein, Analysis module 201 is used to acquire incoming flow data, turbine layout data and terrain data of the target wind farm, input the incoming flow data, turbine layout data and terrain data into the analysis sub-model in the preset wake proxy model, determine the terrain features and airflow distribution law respectively, and adjust the airflow distribution law according to the turbine layout data based on the analysis sub-model to obtain the wake distribution law. The wake proxy model includes analysis sub-model, construction sub-model and prediction sub-model. Construction module 202 is used to input the terrain features, the turbine layout data and the wake distribution pattern into the construction sub-model, determine the global feature vector, node set, edge set, edge feature matrix and node feature matrix of the target wind farm based on the construction sub-model, and establish the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix and node feature matrix; The prediction module 203 is used to input the wind farm map structure into the prediction sub-model, and based on the prediction sub-model, to map the wind farm map structure into the target wind farm prediction output vector set, the prediction output vector set including reference wind speed and power generation.

[0062] The apparatus provided in this disclosure, wherein the analysis module 201 is specifically used for: The terrain data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the terrain elevation of the target wind farm; The unit layout data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the location elevation and airflow distribution influence factors of the target wind farm. The slope amplitude of the target wind farm is calculated based on the terrain elevation and the location elevation. The average slope is calculated based on the slope amplitude. The terrain complexity of the target wind farm is determined by comparing the average slope with a preset grading standard. The terrain features are determined by combining the slope amplitude, average slope, and terrain complexity. The airflow distribution pattern of the target wind farm is determined by analyzing the incoming flow data based on the analysis sub-model. The airflow distribution law is adjusted according to the airflow distribution influencing factor to obtain the wake distribution law.

[0063] The apparatus provided in this disclosure embodiment, wherein the construction module 202 is specifically used for: Based on the constructed sub-model, the nodes and node feature vectors corresponding to each unit in the unit layout data are determined according to the terrain features. Based on the constructed sub-model, the mutual influence relationship between units is determined according to the wake distribution law and the node feature vector, and the edge and edge feature vector of each unit is determined according to the mutual influence relationship; The set of all nodes is the node set, the set of all edges is the edge set, the set of all edge eigenvectors is the edge feature matrix, and the set of all node eigenvectors is the node feature matrix. A global feature vector is determined based on the terrain features and the wake distribution pattern. The wind farm graph structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix.

[0064] The apparatus provided in this disclosure, wherein the prediction module 203 is specifically used for: The prediction sub-model includes a first coding network and a second coding network; Based on the first encoding network, the node feature matrix and the global feature vector are mapped to a first mapping set, which includes the first mapping quantity corresponding to all nodes in the wind farm diagram structure. Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, which includes the second mapping quantities corresponding to all edges in the wind farm graph structure. The predicted output vector set of the target wind farm is determined based on the first mapping value and the second mapping value.

[0065] The apparatus provided in this disclosure, wherein the prediction module 203 is specifically used for: Based on the first encoding network, the node feature vector corresponding to the target node in the node feature matrix and the global feature vector are mapped from the target physical quantity space to a potential space with at least two layers to obtain the first mapping quantity of the target node in the first layer of the potential space. The first mapping set is obtained by combining the first mappings of all nodes.

[0066] The apparatus provided in this disclosure, wherein the prediction module 203 is specifically used for: Based on the second encoding network, the edge feature vector corresponding to the target node is mapped from the target physical quantity space to the potential space to obtain the second mapping quantity of the two nodes associated with the edge corresponding to the edge feature vector; Based on the first mapping quantity and the second mapping quantity corresponding to the two nodes associated with the edge in the first layer, the message intermediate quantity of the first layer is determined. Aggregate the intermediate message quantities of the upstream neighboring nodes corresponding to the target node to obtain an aggregated message, and determine the first mapping quantity of the second layer based on the first mapping quantity of the aggregated message and the target layer; Complete message passing across all layers, and extract the first mapping of the target node in the last layer as the node's potential representation; The latent representation of the node is mapped back to the target physical quantity space to obtain the predicted output vector of the target node; The second mapping set is obtained by taking the second mapping values ​​corresponding to all edge feature vectors; The predicted output vector set is obtained by combining the predicted output vectors corresponding to all nodes in the target wind farm.

[0067] The wind farm wake modeling and prediction device provided in this disclosure can execute the wind farm wake modeling and power generation prediction method provided in any embodiment of this disclosure, and has the corresponding functional modules and beneficial effects of the execution method.

[0068] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program / instruction, which, when executed by a processor, implements the wind farm wake modeling and power generation prediction method in the above embodiments.

[0069] Figure 12 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure.

[0070] The following is a detailed reference. Figure 12 The diagram illustrates a structural schematic suitable for implementing the electronic device in the embodiments of this disclosure. The electronic device in the embodiments of this disclosure may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 12 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0071] like Figure 12As shown, the electronic device may include a processor (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 302 or a program loaded from memory 308 into random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device. The processor 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0072] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 12 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0073] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a memory 308, or installed from a ROM 302. When the computer program is executed by the processor 301, it performs the functions defined in the wind farm wake modeling and power generation prediction method of embodiments of this disclosure.

[0074] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0075] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0076] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0077] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the aforementioned wind farm wake modeling and power generation prediction method.

[0078] Electronic devices can be programmed with computer program code in one or more programming languages ​​or combinations thereof to perform the operations of this disclosure. These programming languages ​​include, but are not limited to, object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as "C" or similar languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0079] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0080] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units are not, in some cases, intended to limit the specific unit.

[0081] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0082] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0083] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

[0084] Furthermore, while the operations are described in a specific order, this should not be construed as requiring these operations to be performed in the specific order shown or in a sequential order. In certain environments, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be construed as limiting the scope of this disclosure. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0085] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

Claims

1. A method for wind farm wake modeling and power generation prediction, characterized in that, include: Acquire the incoming flow data, turbine layout data, and terrain data of the target wind farm. Input the incoming flow data, turbine layout data, and terrain data into the analysis sub-model in the preset wake proxy model to determine the terrain features and airflow distribution patterns respectively. Based on the analysis sub-model, adjust the airflow distribution patterns according to the turbine layout data to obtain the wake distribution patterns. The wake proxy model includes an analysis sub-model, a construction sub-model, and a prediction sub-model. The terrain features, the turbine layout data, and the wake distribution pattern are input into the construction sub-model. Based on the construction sub-model, the global feature vector, node set, edge set, edge feature matrix, and node feature matrix of the target wind farm are determined. The wind farm map structure of the target wind farm is established according to the global feature vector, node set, edge set, edge feature matrix, and node feature matrix. The wind farm map structure is input into the prediction sub-model, and the prediction output vector set that maps the wind farm map structure to the target wind farm is generated based on the prediction sub-model. The prediction output vector set includes reference wind speed and power generation.

2. The method according to claim 1, characterized in that, Acquire inflow data, turbine layout data, and terrain data of the target wind farm. Input the inflow data, turbine layout data, and terrain data into the analysis sub-model of a preset wake proxy model to determine the terrain features and airflow distribution patterns. Based on the analysis sub-model and the turbine layout data, adjust the airflow distribution patterns to obtain the wake distribution patterns, including: The terrain data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the terrain elevation of the target wind farm; The unit layout data is analyzed based on the analysis sub-model in the preset wake proxy model to calculate the location elevation and airflow distribution influence factors of the target wind farm. The slope amplitude of the target wind farm is calculated based on the terrain elevation and the location elevation. The average slope is calculated based on the slope amplitude. The terrain complexity of the target wind farm is determined by comparing the average slope with a preset grading standard. The terrain features are determined by combining the slope amplitude, average slope, and terrain complexity. The airflow distribution pattern of the target wind farm is determined by analyzing the incoming flow data based on the analysis sub-model. The airflow distribution law is adjusted according to the airflow distribution influencing factor to obtain the wake distribution law.

3. The method according to claim 1, characterized in that, The terrain features, turbine layout data, and wake distribution patterns are input into the construction sub-model. Based on the construction sub-model, the global feature vector, node set, edge set, edge feature matrix, and node feature matrix of the target wind farm are determined. The wind farm map structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix, including: Based on the constructed sub-model, the nodes and node feature vectors corresponding to each unit in the unit layout data are determined according to the terrain features. Based on the constructed sub-model, the mutual influence relationship between units is determined according to the wake distribution law and the node feature vector, and the edge and edge feature vector of each unit is determined according to the mutual influence relationship; The set of all nodes is the node set, the set of all edges is the edge set, the set of all edge eigenvectors is the edge feature matrix, and the set of all node eigenvectors is the node feature matrix. A global feature vector is determined based on the terrain features and the wake distribution pattern. The wind farm graph structure of the target wind farm is established based on the global feature vector, node set, edge set, edge feature matrix, and node feature matrix.

4. The method according to claim 1, characterized in that, The wind farm map structure is input into the prediction sub-model, and a set of prediction output vectors mapping the wind farm map structure to the target wind farm is generated based on the prediction sub-model, including: The prediction sub-model includes a first coding network and a second coding network; Based on the first encoding network, the node feature matrix and the global feature vector are mapped to a first mapping set, which includes the first mapping quantity corresponding to all nodes in the wind farm diagram structure. Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, which includes the second mapping quantities corresponding to all edges in the wind farm graph structure. The predicted output vector set of the target wind farm is determined based on the first mapping value and the second mapping value.

5. The method according to claim 4, characterized in that, Based on the first encoding network, mapping the node feature matrix and the global feature vector to a first mapping set includes: Based on the first encoding network, the node feature vector corresponding to the target node in the node feature matrix and the global feature vector are mapped from the target physical quantity space to a potential space with at least two layers to obtain the first mapping quantity of the target node in the first layer of the potential space. The first mapping set is obtained by combining the first mappings of all nodes.

6. The method according to claim 4, characterized in that, Based on the second encoding network, the edge feature vectors are mapped to a second mapping set, and the predicted output vector set of the target wind farm is determined according to the first mapping amount and the second mapping amount, including: Based on the second encoding network, the edge feature vector corresponding to the target node is mapped from the target physical quantity space to the potential space to obtain the second mapping quantity of the two nodes associated with the edge corresponding to the edge feature vector; Based on the first mapping quantity and the second mapping quantity corresponding to the two nodes associated with the edge in the first layer, the message intermediate quantity of the first layer is determined. Aggregate the intermediate message quantities of the upstream neighboring nodes corresponding to the target node to obtain an aggregated message, and determine the first mapping quantity of the second layer based on the first mapping quantity of the aggregated message and the target layer; Complete message passing across all layers, and extract the first mapping of the target node in the last layer as the node's potential representation; The latent representation of the node is mapped back to the target physical quantity space to obtain the predicted output vector of the target node; The second mapping set is obtained by taking the second mapping values ​​corresponding to all edge feature vectors; The predicted output vector set is obtained by combining the predicted output vectors corresponding to all nodes in the target wind farm.

7. A wind farm wake modeling and prediction device, the device comprising: The analysis module is used to acquire the incoming flow data, turbine layout data, and terrain data of the target wind farm. The incoming flow data, turbine layout data, and terrain data are input into the analysis sub-model in the preset wake proxy model to determine the terrain features and airflow distribution patterns. Based on the analysis sub-model, the airflow distribution patterns are adjusted according to the turbine layout data to obtain the wake distribution patterns. The wake proxy model includes an analysis sub-model, a construction sub-model, and a prediction sub-model. The construction module is used to input the terrain features, the turbine layout data and the wake distribution pattern into the construction sub-model, determine the global feature vector, node set, edge set, edge feature matrix and node feature matrix of the target wind farm based on the construction sub-model, and establish the wind farm map structure of the target wind farm according to the global feature vector, node set, edge set, edge feature matrix and node feature matrix; The prediction module is used to input the wind farm map structure into the prediction sub-model, and based on the prediction sub-model, to map the wind farm map structure into a prediction output vector set that represents the target wind farm. The prediction output vector set includes reference wind speed and power generation.

8. An electronic device, characterized in that, include: Memory; processor; as well as Computer programs; The computer program is stored in the memory and configured to be executed by the processor to implement the steps of the method as described in any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, It stores a computer program / instruction thereon, which, when executed by a processor, implements the steps of the method described in any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method according to any one of claims 1-6.

Images